Build a Loan Underwriting Pipeline with LlamaParse

Loan underwriting often involves labor-intensive processes that require analyzing data from diverse financial documents like pay stubs and brokerage statements. In a recent workshop led by Logan, Head of OSS at LlamaIndex, developers constructed a loan underwriting pipeline using LlamaParse tools to automate this process. The application developed during the workshop was capable of extracting structured data from messy financial PDFs and conducting cross-document analysis to generate an underwriting summary. The tech stack employed included async Python, SQLite, FastAPI, Pydantic, and the LlamaCloud SDK, designed to be extensible for future enhancements. The workshop demonstrated three primary services using LlamaParse: converting PDFs to markdown, extracting structured data into Pydantic models, and performing cross-document analysis with a human-in-the-loop approval step. The final service leveraged business-specific knowledge to produce actionable outputs by analyzing multiple documents in combination. The workshop encouraged participants to experiment with the setup by providing a step-by-step guide and repository for hands-on practice.